The AI Research Desk

Today is going to be loud.

The Fed drops its decision this afternoon. At the same time, half the S&P 500 is dumping quarterly earnings into your feed, with profit growth running north of twenty percent and every company spinning its own number into a story. Add the usual river of filings, downgrades, upgrades, and some guy on a podcast yelling about a melt up, and you have a perfect recipe for the thing that quietly wrecks more portfolios than any crash: information overload that turns into bad decisions.

Here is the trap. More information does not make you a better investor. Past a certain point it makes you a worse one, because you start confusing "I read a lot of stuff" with "I understand what matters." You end up reactive, twitchy, and exhausted, which is exactly the state of mind that gets you to sell the bottom and buy the top.

The fix is not to consume more. It is to build a desk that filters. Today I am handing you mine, station by station, so on the loudest day of the month you can walk in, spend twenty minutes, and walk out actually knowing more than the people who spent four hours doom scrolling. Let's build it.

An AI research desk is a small assembly line. Raw information goes in one end, gets captured, summarized, stress tested, and filed, and what comes out the other end is a short daily brief you can actually act on. You build each station once. After that it runs whether you show up or not.

There are four stations plus one rule. Do not skip the rule.

The first failure point is that most people's research lives in seventeen open browser tabs and a vague sense of dread. Nothing is captured, so nothing compounds. Every day you start from zero.

The move is to capture everything the moment it happens, in one place. For any call you sit through, an earnings call, an interview, your own conversation with a partner or advisor, I run Fathom to record and transcribe it automatically and spit out a clean summary with timestamps. The magic is not the recording. It is that six weeks later, when you are trying to remember what a CEO actually promised on the Q2 call versus what he delivered, you have a searchable transcript instead of a fuzzy memory. You build a private library of what people said and when they said it, which is worth its weight in gold when their story conveniently changes.

Rule of thumb: if you would take notes on it, let a tool take better notes than you would, and never lose them.

This is the station that changes everything, and also the one people get most wrong.

The wrong way is to ask an AI "should I buy this stock" and do what it says. That is not research. That is outsourcing your judgment to a very confident autocomplete. The right way is to use AI to compress and pressure test information faster than you ever could alone, then make the call yourself.

I keep Galaxy.ai open because it lets me run several models in one place, so I can put the same question to ChatGPT, Claude, and Grok side by side and watch where they agree and where they don't. When three models built by three different companies all land on the same read, that is signal. When they split, that is my cue to go dig into the disagreement myself. It is like having three sharp analysts who never get tired and never have an agenda.

Here are the three prompts I actually use. Steal them.

The earnings decoder. Paste in an earnings release or transcript and run: "You are a skeptical buy side analyst. Summarize this quarter in ten bullet points. Separate what actually changed from what is spin. Flag any metric that improved only because of an accounting choice, a one time item, or an easy comparison to last year. End with the single biggest risk management is not talking about."

The Fed translator. On a day like today, paste in the FOMC statement and the previous one and run: "Compare these two statements word for word. List every change, however small. For each change, tell me in plain English what it signals about the Fed's next move. Do not editorialize. Just the deltas and what they imply."

The thesis killer. Write out your reason for owning something in a paragraph, then run: "Here is my investment thesis. Your only job is to destroy it. Give me the five strongest arguments for why I am wrong, ranked by how much they should worry me. Assume I am fooling myself." If your thesis survives that, you own it for real. If it folds, you just saved yourself a loss.

Notice that not one of these prompts asks the AI to decide anything. They ask it to compress, compare, and challenge. You are still the one who pulls the trigger. The machine just makes sure you pull it with your eyes open.

Quick worked example so this is not abstract. Say a company reports revenue up 18 percent and the stock pops. You run the earnings decoder and the model flags that 12 of those 18 points came from a single acquisition that closed last quarter, meaning organic growth was more like 6 percent, and it points out that inventory grew faster than sales, which often means a slowdown is coming that management has not admitted yet. Now you are not looking at a headline. You are looking at the actual business. That is a five minute analysis that used to take a trained analyst an hour, and you did it from your kitchen table.

A desk you have to manually operate every morning is a chore, and chores get skipped. So we wire the stations together and let them run before you wake up.

This is where Make.com becomes the spine of the whole operation. You build a scenario, a little visual flowchart of steps, and it fires on a schedule with no code. Mine looks like this:

That is it. By the time I sit down, my brief is written. I read it in ten minutes, add my own notes, and I am done. No tabs. No scrolling. No 90 minute rabbit hole that starts with the Fed and ends with me watching a video about someone's van build. If you have never built a Make scenario, your first one takes an afternoon and every one after that takes twenty minutes, and it pays you back every single morning for the rest of your investing life.

Here is the part almost nobody automates, and it is where a lot of real money actually gets made.

Investing is not only about companies. It is about people, deal flow, who owes you a call, which operator you met at that event who mentioned they were raising. Most of us let those threads rot in a phone we never search. I run that layer through Clay, which quietly keeps my network organized and reminds me who I have gone cold on before an opportunity dies from neglect. Your portfolio is not just your positions. It is your rolodex. Treat it like an asset and it starts behaving like one.

AI in this system is a judgment amplifier, not a judgment replacement. Say it back to yourself.

The models will be confident and wrong sometimes. They will hallucinate a number, misread a footnote, or state a stale fact like gospel. That is fine, as long as you treat every output as a sharp first draft to verify, not a verdict to obey. The second you start auto trading on AI output, or sizing a position because a chatbot sounded sure, you have handed the wheel to something that cannot be held accountable and does not have your money on the line. Never do it. The desk exists to make you faster and clearer. You are still the analyst. You are still the one who signs the ticket.

Here is what an actual morning looks like once the whole thing is humming, so you know what you are building toward.

7:12am. My phone buzzes once. The Make.com scenario finished, the brief is written, and it is sitting in a doc waiting for me. I do not touch it yet. I make coffee like a functional adult.

7:40am. I open the brief. Three overnight items that matter, summarized in plain language, plus two earnings from my watchlist already run through the decoder prompt with the spin stripped out. I read the whole thing in about eight minutes. One earnings report looks better than the headline suggested, so I flag it to look at deeper this weekend. Nothing needs action today. I write two sentences of my own take at the bottom and close the doc.

Total time on research: under fifteen minutes. Total tabs opened: zero. Compare that to the old way, where I would open my feed at 7am "just to check" and surface at 9:30 having learned nothing and absorbed forty conflicting opinions. The desk did not just save me time. It saved me from myself.

That is the real product here. Not speed for its own sake. Calm. A system that hands you the two or three things that matter and refuses to drown you in the rest.

You do not need all four stations live to get value this afternoon. Start with one. Pull today's FOMC statement and the last one, drop them into two models, run the Fed translator prompt, and read the deltas yourself before a single pundit tells you what to think. That one move puts you ahead of ninety percent of people reacting to today's decision, because you read the source and formed your own view first.

Then build Station 1 this week. Then the wiring. Inside a month you have a desk that turns the loudest days of the year into a calm twenty minute routine, and you get your evenings back.

I packaged it all up. You get the Notion template I use to file the daily brief, the exact Make scenario laid out step by step so you can clone it, and all three AI prompts from today formatted so you paste and go. It is the shortcut from cool idea to actually running by Friday.

Want it? Reply with the word DESK and I will send you The Research Desk Blueprint, free. Just reply DESK.

The market will be screaming this afternoon. Everyone else is going to try to drink from the firehose and drown.

You are going to sip from a filtered glass, read the source, and make one or two clear headed decisions instead of forty panicked ones.

That is the desk. Build it once, use it forever.

Friday we go deep on the biggest hidden risk in almost every "diversified" portfolio right now, and how to X-ray your way out of it before the crowd figures it out.

Alex Rivera, Wealth Architect at The Wealth Grid

Wealth is a system, not a guess.